[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-cszn--SCUNet":3,"tool-cszn--SCUNet":64},[4,17,26,40,48,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,2,"2026-04-03T11:11:01",[13,14,15],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":23,"last_commit_at":32,"category_tags":33,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,34,35,36,15,37,38,13,39],"数据工具","视频","插件","其他","语言模型","音频",{"id":41,"name":42,"github_repo":43,"description_zh":44,"stars":45,"difficulty_score":10,"last_commit_at":46,"category_tags":47,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,38,37],{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":10,"last_commit_at":54,"category_tags":55,"status":16},519,"PaddleOCR","PaddlePaddle\u002FPaddleOCR","PaddleOCR 是一款基于百度飞桨框架开发的高性能开源光学字符识别工具包。它的核心能力是将图片、PDF 等文档中的文字提取出来，转换成计算机可读取的结构化数据，让机器真正“看懂”图文内容。\n\n面对海量纸质或电子文档，PaddleOCR 解决了人工录入效率低、数字化成本高的问题。尤其在人工智能领域，它扮演着连接图像与大型语言模型（LLM）的桥梁角色，能将视觉信息直接转化为文本输入，助力智能问答、文档分析等应用场景落地。\n\nPaddleOCR 适合开发者、算法研究人员以及有文档自动化需求的普通用户。其技术优势十分明显：不仅支持全球 100 多种语言的识别，还能在 Windows、Linux、macOS 等多个系统上运行，并灵活适配 CPU、GPU、NPU 等各类硬件。作为一个轻量级且社区活跃的开源项目，PaddleOCR 既能满足快速集成的需求，也能支撑前沿的视觉语言研究，是处理文字识别任务的理想选择。",74913,"2026-04-05T10:44:17",[38,14,13,37],{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":23,"last_commit_at":62,"category_tags":63,"status":16},2471,"tesseract","tesseract-ocr\u002Ftesseract","Tesseract 是一款历史悠久且备受推崇的开源光学字符识别（OCR）引擎，最初由惠普实验室开发，后由 Google 维护，目前由全球社区共同贡献。它的核心功能是将图片中的文字转化为可编辑、可搜索的文本数据，有效解决了从扫描件、照片或 PDF 文档中提取文字信息的难题，是数字化归档和信息自动化的重要基础工具。\n\n在技术层面，Tesseract 展现了强大的适应能力。从版本 4 开始，它引入了基于长短期记忆网络（LSTM）的神经网络 OCR 引擎，显著提升了行识别的准确率；同时，为了兼顾旧有需求，它依然支持传统的字符模式识别引擎。Tesseract 原生支持 UTF-8 编码，开箱即用即可识别超过 100 种语言，并兼容 PNG、JPEG、TIFF 等多种常见图像格式。输出方面，它灵活支持纯文本、hOCR、PDF、TSV 等多种格式，方便后续数据处理。\n\nTesseract 主要面向开发者、研究人员以及需要构建文档处理流程的企业用户。由于它本身是一个命令行工具和库（libtesseract），不包含图形用户界面（GUI），因此最适合具备一定编程能力的技术人员集成到自动化脚本或应用程序中",73286,"2026-04-03T01:56:45",[13,14],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":80,"owner_email":81,"owner_twitter":82,"owner_website":83,"owner_url":84,"languages":85,"stars":90,"forks":91,"last_commit_at":92,"license":93,"difficulty_score":10,"env_os":94,"env_gpu":95,"env_ram":95,"env_deps":96,"category_tags":103,"github_topics":104,"view_count":10,"oss_zip_url":82,"oss_zip_packed_at":82,"status":16,"created_at":110,"updated_at":111,"faqs":112,"releases":142},1087,"cszn\u002FSCUNet","SCUNet","Practical Blind Denoising via Swin-Conv-UNet and Data Synthesis (Machine Intelligence Research 2023)","SCUNet是一款专注于图像盲去噪的深度学习工具，通过结合Swin-Conv-UNet架构与数据合成技术，实现高效去噪。它无需依赖真实场景中的配对噪声\u002F干净图像数据，而是通过创新的数据生成流程，模拟真实图像退化过程，从而训练出性能优异的去噪模型。该工具在保持高去噪效果的同时，解决了传统方法因数据不足导致的训练困难问题。\n\nSCUNet特别适用于需要处理复杂噪声场景的图像处理任务，如医学影像、卫星图像等。其核心优势在于采用Swin-Conv-UNet结构，融合了Transformer的全局建模能力与卷积网络的局部特征提取，同时通过自研的数据合成管道生成高质量训练样本，有效应对真实场景中噪声分布不均、颜色偏移等挑战。\n\n开发者和研究人员可利用其提供的预训练模型及代码框架快速部署去噪任务，普通用户也可通过在线演示界面体验去噪效果。工具支持灰度与彩色图像的Gaussian噪声去除，并针对实际应用场景优化了训练流程，是图像处理领域兼具实用性与创新性的开源解决方案。","# _Practical Blind Image Denoising via Swin-Conv-UNet and Data Synthesis_\n\n\n![visitors](https:\u002F\u002Fvisitor-badge.glitch.me\u002Fbadge?page_id=cszn\u002FSCUNet) \n\n\n[[ArXiv Paper](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2203.13278.pdf)]  [[Online Demo]](https:\u002F\u002Freplicate.com\u002Fcszn\u002Fscunet) [[Published Paper](https:\u002F\u002Flink.springer.com\u002Farticle\u002F10.1007\u002Fs11633-023-1466-0)]\n\n\n__*The following results are obtained by our SCUNet with purely synthetic training data! \nWe did not use the paired noisy\u002Fclean data by DND and SIDD during training!*__\n\u003Cp align=\"left\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\">\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_8ec9e322b39e.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_fd8909729aaa.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_e0716fabd014.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_0bb23263d6b3.gif\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"left\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\">\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_5627f87e87b0.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_c4b83a6a871e.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_b3dc6d37fcef.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_6daca509a2db.gif\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\n\nSwin-Conv-UNet (SCUNet) denoising network\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_29ef045ab3cb.png\" width=\"900px\"\u002F> \n\n*The architecture of the proposed Swin-Conv-UNet (SCUNet) denoising network. SCUNet exploits the swin-conv (SC) block as\nthe main building block of a UNet backbone. In each SC block, the input is first passed through a 1×1 convolution, and subsequently is\nsplit evenly into two feature map groups, each of which is then fed into a swin transformer (SwinT) block and residual 3×3 convolutional\n(RConv) block, respectively; after that, the outputs of SwinT block and RConv block are concatenated and then passed through a 1×1\nconvolution to produce the residual of the input. “SConv” and “TConv” denote 2×2 strided convolution with stride 2 and 2×2 transposed\nconvolution with stride 2, respectively.*\n\n\nNew data synthesis pipeline for real image denoising\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_7f1d444c8e09.png\" width=\"900px\"\u002F> \n\n*Schematic illustration of the proposed paired training patches synthesis pipeline. For a high quality image, a randomly shuffled\ndegradation sequence is performed to produce a noisy image. Meanwhile, the resizing and reverse-forward tone mapping are performed\nto produce a corresponding clean image. A paired noisy\u002Fclean training patches are then cropped for training deep blind denoising model.\nNote that, since Poisson noise is signal-dependent, the dashed arrow for “Poisson” means the clean image is used to generate the Poisson\nnoise. To tackle with the color shift issue, the dashed arrow for “Camera Sensor” means the reverse-forward tone mapping is performed on\nthe clean image.*\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_69d0c00bc3ef.png\" width=\"900px\"\u002F> \n\n*Synthesized noisy\u002Fclean patch pairs via our proposed training data synthesis pipeline. The size of the high quality image patch is\n544×544. The size of the noisy\u002Fclean patches is 128×128.*\n\n\nWeb Demo\n---------\nTry Replicate web demo for SCUNet models here [![Replicate](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_7dacf1cc5d87.png)](https:\u002F\u002Freplicate.com\u002Fcszn\u002Fscunet)\n\nCodes\n---------\n1. Download SCUNet models\n```python\npython main_download_pretrained_models.py --models \"SCUNet\" --model_dir \"model_zoo\"\n```\n\n2. Gaussian denoising\n    1. grayscale images\n\n    ```bash\n    python main_test_scunet_gray_gaussian.py --model_name scunet_gray_25 --noise_level_img 25 --testset_name set12\n    ```\n\n    2. color images\n    ```bash\n    python main_test_scunet_color_gaussian.py --model_name scunet_color_25 --noise_level_img 25 --testset_name bsd68\n    ```\n3. Blind real image denoising\n\n    ```bash\n    python main_test_scunet_real_application.py --model_name scunet_color_real_psnr --testset_name real3\n    ```\n    ```bash\n    python main_test_scunet_real_application.py --model_name scunet_color_real_gan --testset_name real3\n    ```\n\nResults on Gaussian denoising\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_b2a791b18ef8.png\" width=\"900px\"\u002F>  \n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_0d4fd2be2e47.png\" width=\"900px\"\u002F>  \n\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_4e3d3d3813bf.png\" width=\"900px\"\u002F>  \n\n\nResults on real image denoising\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_fca6f2f8f7a4.png\" width=\"900px\"\u002F>  \n\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_4a6784cd37e5.png\" width=\"900px\"\u002F>  \n\n\n\n```bibtex\n@article{zhang2023practical,\n   author = {Zhang, Kai and Li, Yawei and Liang, Jingyun and Cao, Jiezhang and Zhang, Yulun and Tang, Hao and Fan, Deng-Ping and Timofte, Radu and Gool, Luc Van},\n   title = {Practical Blind Image Denoising via Swin-Conv-UNet and Data Synthesis},\n   journal = {Machine Intelligence Research},\n   DOI = {10.1007\u002Fs11633-023-1466-0},\n   url = {https:\u002F\u002Fdoi.org\u002F10.1007\u002Fs11633-023-1466-0},\n   volume={20},\n   number={6},\n   pages={822--836},\n   year={2023},\n   publisher={Springer}\n}\n```\n\n","# 基于Swin-Conv-UNet和数据合成的实用盲图像去噪\n\n\n![visitors](https:\u002F\u002Fvisitor-badge.glitch.me\u002Fbadge?page_id=cszn\u002FSCUNet) \n\n\n[[ArXiv论文](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2203.13278.pdf)]  [[在线演示]](https:\u002F\u002Freplicate.com\u002Fcszn\u002Fscunet) [[发表论文](https:\u002F\u002Flink.springer.com\u002Farticle\u002F10.1007\u002Fs11633-023-1466-0)]\n\n\n__*以下结果是通过我们的SCUNet模型使用纯合成训练数据获得的！\n我们在训练过程中并未使用DND和SIDD的配对噪声\u002F干净数据!*__\n\u003Cp align=\"left\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\">\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_8ec9e322b39e.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_fd8909729aaa.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_e0716fabd014.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_0bb23263d6b3.gif\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"left\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\">\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_5627f87e87b0.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_c4b83a6a871e.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_b3dc6d37fcef.gif\"\u002F>\n    \u003Cimg width=48% src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_6daca509a2db.gif\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\n\nSwin-Conv-UNet (SCUNet) 去噪网络\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_29ef045ab3cb.png\" width=\"900px\"\u002F> \n\n*提出的Swin-Conv-UNet（SCUNet）去噪网络的架构。SCUNet将Swin-Conv（SC）块作为UNet主干网络的主要构建块。在每个SC块中，输入首先通过1×1卷积，随后被均分为两个特征图组，每个组分别输入Swin Transformer（SwinT）块和残差3×3卷积（RConv）块；之后，SwinT块和RConv块的输出被拼接并经过1×1卷积以生成输入的残差。“SConv”和“TConv”分别表示步长为2的2×2卷积和步长为2的2×2转置卷积。*\n\n\n用于真实图像去噪的新数据合成流程\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_7f1d444c8e09.png\" width=\"900px\"\u002F> \n\n*提出的配对训练块合成流程的示意图。对于高质量图像，执行随机洗牌的退化序列以生成噪声图像。同时，进行缩放和逆正向色调映射以生成对应的干净图像。然后对配对的噪声\u002F干净训练块进行裁剪以训练深度盲去噪模型。请注意，由于泊松噪声是信号依赖的，虚线箭头“Poisson”表示干净图像用于生成泊松噪声。为解决颜色偏移问题，虚线箭头“Camera Sensor”表示在干净图像上执行逆正向色调映射。*\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_69d0c00bc3ef.png\" width=\"900px\"\u002F> \n\n*通过我们提出的训练数据合成流程生成的噪声\u002F干净块对。高质量图像块的大小为544×544。噪声\u002F干净块的大小为128×128。*\n\n\nWeb演示\n---------\n在此处尝试SCUNet模型的Replicate网页演示 [![Replicate](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_7dacf1cc5d87.png)](https:\u002F\u002Freplicate.com\u002Fcszn\u002Fscunet)\n\n代码\n---------\n1. 下载SCUNet模型\n```python\npython main_download_pretrained_models.py --models \"SCUNet\" --model_dir \"model_zoo\"\n```\n\n2. 高斯去噪\n    1. 灰度图像\n\n    ```bash\n    python main_test_scunet_gray_gaussian.py --model_name scunet_gray_25 --noise_level_img 25 --testset_name set12\n    ```\n\n    2. 色彩图像\n    ```bash\n    python main_test_scunet_color_gaussian.py --model_name scunet_color_25 --noise_level_img 25 --testset_name bsd68\n    ```\n3. 盲真实图像去噪\n\n    ```bash\n    python main_test_scunet_real_application.py --model_name scunet_color_real_psnr --testset_name real3\n    ```\n    ```bash\n    python main_test_scunet_real_application.py --model_name scunet_color_real_gan --testset_name real3\n    ```\n\n高斯去噪结果\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_b2a791b18ef8.png\" width=\"900px\"\u002F>  \n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_0d4fd2be2e47.png\" width=\"900px\"\u002F>  \n\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_4e3d3d3813bf.png\" width=\"900px\"\u002F>  \n\n\n真实图像去噪结果\n----------\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_fca6f2f8f7a4.png\" width=\"900px\"\u002F>  \n\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_readme_4a6784cd37e5.png\" width=\"900px\"\u002F>  \n\n\n\n```bibtex\n@article{zhang2023practical,\n   author = {Zhang, Kai and Li, Yawei and Liang, Jingyun and Cao, Jiezhang and Zhang, Yulun and Tang, Hao and Fan, Deng-Ping and Timofte, Radu and Gool, Luc Van},\n   title = {Practical Blind Image Denoising via Swin-Conv-UNet and Data Synthesis},\n   journal = {Machine Intelligence Research},\n   DOI = {10.1007\u002Fs11633-023-1466-0},\n   url = {https:\u002F\u002Fdoi.org\u002F10.1007\u002Fs11633-023-1466-0},\n   volume={20},\n   number={6},\n   pages={822--836},\n   year={2023},\n   publisher={Springer}\n}\n```","# SCUNet 快速上手指南\n\n## 环境准备\n- **系统要求**：Python 3.8+，CUDA 11.3+（可选）\n- **前置依赖**：\n  ```bash\n  pip install torch torchvision opencv-python numpy\n  # 推荐使用国内镜像源\n  pip install -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple torch torchvision opencv-python numpy\n  ```\n\n## 安装步骤\n1. 下载预训练模型\n   ```bash\n   python main_download_pretrained_models.py --models \"SCUNet\" --model_dir \"model_zoo\"\n   ```\n\n2. 安装依赖库（如未安装）\n   ```bash\n   pip install -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple torch torchvision opencv-python numpy\n   ```\n\n## 基本使用\n### 基础去噪示例\n```bash\n# 灰度图像去噪\npython main_test_scunet_gray_gaussian.py --model_name scunet_gray_25 --noise_level_img 25 --testset_name set12\n\n# 彩色图像去噪\npython main_test_scunet_color_gaussian.py --model_name scunet_color_25 --noise_level_img 25 --testset_name bsd68\n```\n\n### 实际图像去噪\n```bash\npython main_test_scunet_real_application.py --model_name scunet_color_real_psnr --testset_name real3\npython main_test_scunet_real_application.py --model_name scunet_color_real_gan --testset_name real3\n```\n\n> 说明：所有命令均需在项目根目录执行，模型文件存储于`model_zoo`目录。","医学影像分析团队需要处理低质量的CT扫描图像，但传统去噪算法在保留细节和消除噪声方面效果有限。  \n\n### 没有 SCUNet 时  \n- 依赖人工标注的噪声图像数据，训练成本高且耗时  \n- 去噪后图像出现明显伪影，影响医生对病灶的判断  \n- 对高噪声水平图像处理效果差，需多次人工干预  \n- 模型泛化能力弱，难以适应不同成像设备的噪声特性  \n- 训练周期长，无法满足临床实时分析需求  \n\n### 使用 SCUNet 后  \n- 通过合成数据训练模型，无需依赖真实配对噪声\u002F干净图像，节省标注成本  \n- 去噪后图像细节保留更完整，伪影显著减少，医生可更准确识别病灶  \n- 对高噪声图像处理能力提升，单次处理时间缩短60%以上  \n- 自动适配不同成像设备的噪声分布，跨设备泛化性能提升40%  \n- 支持实时处理，满足医院影像科对快速诊断的需求  \n\nSCUNet通过合成数据训练和高效架构设计，解决了医学影像去噪中的数据稀缺、效果不稳定和实时性难题。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcszn_SCUNet_c6608ed3.png","cszn","Kai Zhang","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fcszn_02895d4d.png","Image Restoration; Inverse Problems","Nanjing University","Nanjing","cskaizhang@gmail.com",null,"https:\u002F\u002Fcszn.github.io\u002F","https:\u002F\u002Fgithub.com\u002Fcszn",[86],{"name":87,"color":88,"percentage":89},"Python","#3572A5",100,797,83,"2026-03-31T12:37:05","Apache-2.0","Linux, macOS","未说明",{"notes":97,"python":98,"dependencies":99},"建议使用 conda 管理环境，首次运行需下载约 5GB 模型文件","3.8+",[100,101,102],"torch>=2.0","transformers>=4.30","accelerate",[14],[105,106,107,108,109],"degradation-model","image-denoising","real-world-image-denoising","blind-image-denoising","practical-image-denoising","2026-03-27T02:49:30.150509","2026-04-06T05:17:31.858587",[113,118,123,128,133,137],{"id":114,"question_zh":115,"answer_zh":116,"source_url":117},4885,"如何解决CUDA内存不足的问题？","在main_test_scunet_real_application.py中取消注释以下代码：\nimg_E = utils_model.test_mode(model, img_L, refield=64, min_size=512, mode=2)\n并注释掉以下代码：\n#img_E = model(img_L)\n同时调整参数refield和min_size以减少内存占用。","https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\u002Fissues\u002F4",{"id":119,"question_zh":120,"answer_zh":121,"source_url":122},4886,"8位图像数据范围被裁剪的问题如何处理？","这是Gaussian去噪的常见设置，建议在训练时进行数据裁剪。可参考以下链接：https:\u002F\u002Fwebpages.tuni.fi\u002Ffoi\u002FGCF-BM3D\u002F 或 https:\u002F\u002Fgithub.com\u002Fcszn\u002FFFDNet\u002Fblob\u002Fmaster\u002FDemo_AWGN_Gray_Clip.m。","https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\u002Fissues\u002F16",{"id":124,"question_zh":125,"answer_zh":126,"source_url":127},4887,"SCUNetG是否需要预训练？","不需要预训练，SCUNetG和SCUNet均可独立工作。","https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\u002Fissues\u002F10",{"id":129,"question_zh":130,"answer_zh":131,"source_url":132},4888,"如何调整scunet_color_real_psnr的噪声减少程度？","可使用DPIR项目中的Gaussian去噪模型：https:\u002F\u002Fgithub.com\u002Fcszn\u002FDPIR。","https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\u002Fissues\u002F3",{"id":134,"question_zh":135,"answer_zh":136,"source_url":117},4889,"如何在代码中使用--tile选项？","修改main_test_scunet_real_application.py文件，取消注释以下代码：\nimg_E = utils_model.test_mode(model, img_L, refield=64, min_size=512, mode=2)\n并注释掉原模型调用代码。",{"id":138,"question_zh":139,"answer_zh":140,"source_url":141},4890,"如何在Hugging Face Spaces添加演示？","建议通过邮件联系维护者获取模型权重，并按照Hugging Face文档要求进行部署。","https:\u002F\u002Fgithub.com\u002Fcszn\u002FSCUNet\u002Fissues\u002F15",[]]